Google’s Home MCP lets a personal AI agent read a household’s devices and run commands on them, with lights, thermostats and vacuums among the device categories its documentation names. It stops short of doors. Google’s developer page says the server prohibits sensitive actions such as unlocking them, and a release note that describes access to “all of the devices” has to be read against that carve-out.
The feature entered early access on September 16, 2026, and it is narrower than the one-line summary suggests in three further ways: experimental traits, no automations, and a paid-subscription gate.
Google’s one-sentence release note
Google’s Google Home release notes list the item under the Google Home Platform heading for September 16, 2026. It says the company is beginning the rollout of the new Google Home MCP in early access, and that once it is available a personal agent, with Antigravity and OpenClaw given as examples, can interact with all of the devices and event history in the Google Home ecosystem.
MCP stands for Model Context Protocol, an open standard that lets AI models and chatbots fetch data from apps and services and act on them. Google also pointed readers of the note to a longer announcement post on the Google Home and Nest Community, which carries the product’s formal introduction as a way for an agent to interact with a home. The release note itself is the controlling statement, and it is a single sentence.
The device classes the documentation names
The developer page for the Google Home MCP server does not publish an exhaustive device list, but its category and device-type vocabulary is concrete. It names lighting, climate and thermostats, safety and security with locks, media with televisions, household appliances with vacuums, kitchen appliances with coffee makers, bed and bath with showers, and outdoor equipment with sprinklers. The page also refers to cameras and doorbells in its section on familiar-face consent.
Five tools carry the work: list_homes, list_home_resources, list_home_states, list_home_history and run_home_actions. The first four read, covering the homes an account can see, the devices and rooms in them, their current states and their event history. The fifth executes parameterized action commands on target devices. In practice, that structure means an agent can ask which lights are on and then switch them off, or read a thermostat and change its setpoint, with every command passing through the same server and the same safety filter.
Android Authority’s coverage describes the same ground from the user side: monitoring activity, controlling devices and reaching camera feeds through agents including Claude, Google’s Antigravity, OpenClaw and Hermes, for American users on a Google Home Premium Advanced subscription.
The carve-outs in Google’s developer documentation
The docs state the limit directly: “Home MCP enforces rate limits and safety protections, such as prohibiting sensitive actions like unlocking doors.” The lock category appears in the vocabulary, so an agent can see that a lock exists, but unlocking is the example Google chose for what stays off the table. Door unlocking is the case the documentation spells out, and the wording “such as” leaves room for other sensitive actions the page does not list.
Two more limits sit on the same page. Experimental traits are included to widen what an agent can do, and Google warns that they may not behave as expected, directing problems to a public issue tracker. And automations are out: the documentation says creating and managing them with Home MCP is not supported at this time, with support planned for a future release.
The Home MCPs overview adds the design intent, that the server is built to Google’s security and privacy standards so agents cannot take sensitive actions on a user’s behalf. It also separates the consumer server from a second, Home Developer MCP that serves people building integrations, with documentation on Matter, OpenThread and cloud-to-cloud requirements. The Home MCPs overview lists the consumer server as using a user-authorized OAuth flow.
Premium Advanced, Google Cloud and a United States launch
Access requires an active Google Home Premium Advanced subscription and access to a Google Cloud project, and the feature is labeled early access. The Next Web’s report on the launch puts availability at the United States only, in English, and notes that setup involves a Google Cloud project, an OAuth consent screen and client credentials, a chore Google suggests handing to the agent itself. Its coverage of the launch also quotes the rate-limit sentence from Google’s docs.
The gap between “all of the devices” in the release note and the documentation’s exclusions is the line to read closely. The note describes reach across the Google Home ecosystem, devices and event history alike, while the developer page describes what that reach may do once an agent arrives: switch lights, adjust thermostats and run vacuums, but not open a door, and not yet build an automation.
This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.
More from Morning Overview
- Invisible volcanic gas at a California ski resort has killed three ski patrollers
- Verizon just moved to cut landline phone service in 9 states within weeks
- Tropical Storm Rachel is dumping up to 12 inches on four Mexican states on its way to major hurricane strength
- The NSA is again telling phone owners to switch off one location setting
